How to Read a Canine Cancer Study: A Plain-Language Guide to the Numbers.

Part of our complete guide for newly diagnosed dogs.

A number is not a verdict

When a dog is diagnosed with cancer, the numbers arrive fast: a survival time, a response rate, a percentage. They feel final. They are not. Every one of those numbers came from a study of other dogs, and how much it tells you about your dog depends entirely on how the study was built and how the number is worded.

This is not a call to distrust your oncologist. A good oncologist is the best guide you have. It is a call to understand the same numbers they use, so the conversation is a real one. The rest of this guide walks through the handful of ideas that do almost all the work.

"Median" means the middle, not your dog

Most survival numbers you will meet are a median. A median is simply the middle of a line-up. If you line up every dog in a study from the shortest survival to the longest, the median is the dog standing in the exact middle. Half the dogs lived less than that; half lived more.

That last part is the part people miss. A median survival of, say, 1.6 months does not mean dogs live 1.6 months. It means that at 1.6 months, half were still alive, and some of those went on to live far longer. In one of the largest studies of splenic hemangiosarcoma treated with surgery alone, according to PubMed, the median survival was about 1.6 months across 208 dogs (Wendelburg et al., Journal of the American Veterinary Medical Association, 2015). Read as a verdict, that number is crushing. Read correctly, it is the midpoint of a wide spread, and your dog's place in that spread is not yet written.

Why two honest studies disagree

Search any dog cancer long enough and you will find two trustworthy studies that give different numbers. This is not a sign that someone is wrong. It usually means the two studies enrolled different dogs.

Take that same question, survival after removing the spleen for hemangiosarcoma. One study of 208 dogs reports about 1.6 months (Wendelburg 2015). An older study of 59 dogs reports about 86 days, closer to three months (Kim et al., JAVMA, 2007). Both are real, peer-reviewed results. The honest way to hold them is not to pick a winner but to state the range: surgery alone buys somewhere in the region of one and a half to three months, depending on the mix of dogs and disease stages a given study happened to enroll. A range that carries its studies is more honest than a single confident number, and you should be slightly suspicious of anyone who gives you only the single number.

Cohort size: why the number of dogs matters

The number of animals in a study, written as n, is one of the fastest quality checks you can do. A result from 8 dogs and a result from 800 dogs are not the same kind of fact, even when the percentage looks identical.

Small studies swing hard on luck. If a treatment is tested in 8 dogs and 4 respond, that is "50 percent," but a single dog going the other way would have made it 37 percent or 63 percent. Larger studies are steadier, because one unusual dog cannot move the average much. This is why, when two numbers conflict, the one built on more animals usually deserves more weight, all else being equal. It is also why a striking result from a tiny study is a reason to be interested, not yet a reason to be convinced.

The evidence ladder: not all studies are equal

The single most useful habit is to ask what kind of study a number came from. Study designs sit on a ladder, from weakest to strongest evidence:

• Case report (n = 1). One dog, described in detail. It can prove a treatment is possible and worth studying. It cannot tell you how often it works.
• Retrospective study. Researchers look backward through records of dogs already treated. Useful and common in veterinary medicine, but the dogs were not assigned at random, so hidden differences between groups can quietly shape the result.
• Prospective study. Researchers decide the plan in advance and then follow dogs forward. Cleaner than looking backward, because the questions are set before the answers are known.
• Randomised controlled trial (RCT). Dogs are assigned by chance to the treatment or to a comparison group. Randomising is what lets you credit the treatment for a difference, rather than some accident of which dogs ended up where. This is the strongest common design, and it is rare in dog cancer because it is expensive and hard.

A real example of the top rung: the drug Palladia (toceranib) for mast cell tumours was tested in a randomised, double-blind, placebo-controlled trial, and dogs on the drug had an objective response rate of 37 percent versus 8 percent on placebo, across 145 dogs (London et al., Clinical Cancer Research, 2009). Because it was randomised and placebo-controlled, that gap is strong evidence the drug did something. Not every number you meet is built that well, and knowing the difference is most of the skill.

Historical controls, and why they flatter

Some studies compare a treated group not against a group treated at the same time, but against historical controls: dogs treated in the past, whose records are used as the comparison. This is often the only practical option, but it tilts the deck. Care improves over the years, the treated dogs are chosen in the present, and the two groups were never really alike. Comparisons against historical controls tend to make a treatment look better than a head-to-head trial would.

The canine melanoma vaccine ONCEPT is the textbook case. The study behind its licensure compared 58 vaccinated dogs against 53 historical controls and reported longer survival, and its first author worked for the manufacturer (Grosenbaugh et al., American Journal of Veterinary Research, 2011). A later retrospective review of 45 dogs found no improvement in survival at all (Ottnod et al., Veterinary and Comparative Oncology, 2013). Neither study was a randomised controlled trial, which is the honest reason the debate is still open. When a result rests on historical controls, treat it as a promising lead, not a settled fact.

Relative versus absolute: the "44 percent" trap

This is the trap that catches the most people, in dog and human medicine alike. A headline says a treatment "cut the risk by 44 percent." Cut it from what, to what?

"44 percent" is usually a relative reduction. If a risk drops from 40 out of 100 down to about 22 out of 100, that is a 44 percent relative cut, but the absolute change is 18 fewer dogs in 100. Relative numbers sound bigger. Both can be true at once, and the honest write-up gives you both. The much-discussed human melanoma vaccine mRNA-4157 plus Keytruda is described this way: it lowered the risk of recurrence or death by about 44 percent versus Keytruda alone in a randomised phase 2b trial (Weber et al., Lancet, 2024). That is genuinely encouraging, and it is a relative figure from a mid-stage human trial, not a finished result in dogs. Holding both facts at once is exactly the skill.

What a p-value does and does not tell you

You will sometimes see a p-value, written as something like "p = 0.03." It is a rough answer to one narrow question: if the treatment truly did nothing, how likely is a result at least this lopsided by chance alone? A small p-value (by convention, under 0.05) is taken as a signal that the result is probably not a fluke.

Two cautions. First, a p-value does not tell you how big or how important an effect is, only how likely it is to be noise. A tiny, meaningless difference can be "statistically significant" in a huge study. Second, the 0.05 line is a convention, not a law of nature. In that same human vaccine trial, the headline recurrence result carried a p-value of 0.053, just over the usual line (Weber et al., 2024). That is why it is called a promising phase 2b result with a larger trial underway, rather than a closed case. Numbers near the line deserve the word "promising," not "proven."

Stage can matter more than treatment

Finally, the biggest driver of a survival number is often not the treatment at all. It is stage: how far the cancer had already spread when it was found. Comparing two treatments without matching stage is like comparing two runners without mentioning that one started halfway down the track.

In that 59-dog hemangiosarcoma study, according to PubMed, median survival was 345 days for Stage I, 93 days for Stage II, and 68 days for Stage III (Kim et al., 2007). A dog caught while the cancer was still confined lived roughly five times longer than one caught after it had spread, and no drug in the study moved the needle nearly that far. When you read a prognosis, always ask which stage it describes, and which stage your dog is.

A checklist for any number you are given

When your vet or an article gives you a number, these six questions turn it from a verdict into information:

1. Is it a median? If so, remember half the dogs did better, some by a lot.
2. How many dogs? A bigger n is steadier than a small one.
3. What kind of study? Case report, retrospective, prospective, or randomised trial. Higher on the ladder means stronger evidence.
4. Compared against what? A concurrent control group is stronger than historical controls.
5. Relative or absolute? Ask what the percentage moved from, and to.
6. Which stage, and which cancer subtype? The same disease at a different stage is almost a different question.

You do not need to interrogate your oncologist. But holding these questions in mind lets you hear what they say more clearly, and ask the one or two that matter for your dog.

How we handle numbers here

Because these traps are everywhere, we hold our own writing to a simple rule: every survival or response number we publish carries its source, its cohort size, and its study design, linked to the original research. We do not quote a statistic from another advocacy site; we trace it to the study that first reported it. Where the evidence is weak or the studies disagree, we say so and give the range. You can see the rule applied in our cited cancer guides, for example the hemangiosarcoma guide and the deeper piece on what a median survival really means.

Sources

1. Wendelburg et al., JAVMA, 2015 (splenic hemangiosarcoma; retrospective, 208 dogs). PubMed · DOI
2. Kim et al., JAVMA, 2007 (splenic hemangiosarcoma, survival by stage; 59 dogs). PubMed · DOI
3. London et al., Clinical Cancer Research, 2009 (toceranib/Palladia; randomised, double-blind, placebo-controlled, 145 dogs). PubMed · DOI
4. Grosenbaugh et al., American Journal of Veterinary Research, 2011 (ONCEPT; 58 vaccinated vs 53 historical controls; manufacturer-affiliated first author). PubMed · DOI
5. Ottnod et al., Veterinary and Comparative Oncology, 2013 (ONCEPT; retrospective, 45 dogs). PubMed · DOI
6. Weber et al., Lancet, 2024 (KEYNOTE-942; randomised phase 2b, 157 human patients). PubMed · DOI